18
Elif Ertekin Department of Mechanical Science & Engineering, University of Illinois, Urbana-Champaign Materials Research Society - Fall 2019 - Boston MA - 1 Dec 2019 Density Functional Theory and Electronic Structure Calculations Tutorial EL04 - Introduction to Chalcogenide Discovery and Design Today’s goal is to introduce how density functional theory can be used to predict the bulk and defect properties of chalcogenides. photovoltaics thermoelectrics Hot Cold p-type

Density Functional Theory and Electronic Structure ...jaramillo.mit.edu/sites/default/files/2019 Fall MRS - Ertekin - DFT.pdf · Density Functional Theory and Electronic Structure

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

  • Elif Ertekin Department of Mechanical Science & Engineering,

    University of Illinois, Urbana-Champaign

    Materials Research Society - Fall 2019 - Boston MA - 1 Dec 2019

    Density Functional Theory and Electronic Structure Calculations

    Tutorial EL04 - Introduction to Chalcogenide Discovery and Design

    Today’s goal is to introduce how density functional theory can be used to predict the bulk and defect properties of chalcogenides.

    photovoltaics thermoelectrics

    Hot

    Cold

    p-type

  • what can be reliably calculated?

    what needs more scrutiny?

    ground vs. excited state?

    equilibrium vs. non-equilibrium?

    what properties are we interested in?

    We want to be able to predict the properties of chalcogenides (and all materials) accurately and effectively on a computer.

    Crystal structure-------------------------------------• Structure and total energies

    of compounds• Lattice constants (at T = 0 K) • Other properties derived

    from crystal structure



    Thermodynamic Stability

    -------------------------------------- Phase stabilities - Composition space vs.

    chemical potential space

    Defects & Doping-------------------------------------- Calculation of defect formation energies- Translate to defect concentrations, carrier concentrations, dopability



    Band Structure-------------------------------------- Dispersion spectrum for a

    compound- Band gap, quasi-particle

    energies, optical properties - Transport: electron and hole

    effective masses



  • The Schrodinger equation describes all properties of all materials.

    H = − 1

    2∇2 −∑

    Znre−n

    +1re−e

    Hψ r1,..., rn( ) = Eψ r1,..., rn( )

    Non-interac+ngpart

    ContainsanElectron-Electron

    Distance!

    •The wave function encodes all information about a system of particles (here, electrons and nuclei)

    •Many approximations already: non-relativistic, time-independent, Born-Oppenheimer.

    • Eigenvalue/eigenfunction problem: wave functions are eigenfunctions and energies are the eigenvalues

    wave functionenergy

    “Where did we get that (equation) from? Nowhere. It is not possible to derive it from anything you know. It came out of the mind of Schrödinger.” - Richard Feynman

    It is not possible to solve exactly the Schrodinger equation except in the simplest systems.

    •Electron-electron term makes the solution non-separable

    •For an n electron system, wave function is a function of 3n variables

    •Quickly becomes unmanageable, e.g. storing a wave function on a 2x2x2 real space grid requires storing

    •1 electron : 8 values •10 electrons : 109 values •100 electrons : 1090 values •1000 electrons : 10900 values

    H = − 1

    2∇2 −∑

    Znre−n

    +1re−e

    Hψ r1,..., rn( ) = Eψ r1,..., rn( )

    Non-interac+ngpart

    ContainsanElectron-Electron

    Distance!

    wave functionenergy

  • Tradeoffs: Today DFT represents the best available compromise between accuracy and computational efficiency.

    inspired by Lucas Wagner, UIUC

    DFT works by mapping the interacting Schrodinger equation onto an effective non-interacting equation.

    •Founded on the two Hohenberg-Kohn theorems

    •Hohenberg-Kohn I: one to one mapping between the external potential of the nuclei, ground state wave function, and ground state charge density

    •Hohenberg-Kohn II: there exists a universal functional of the density F[ρ(r)] such that the ground state energy E is minimized at the true ground state density

    •Typically formulated as an Euler-Lagrange system of equations

    E ρ(r)[ ] = Vext∫ (r)ρ(r)dr + F ρ(r)[ ]

    UNIVERSALbut unknown

    Hohenberg and Kohn, 1964

  • The Kohn-Sham approach is most widely used in practice (I).

    E = Ψ*(Vext + T +Vint )Ψ∫ d3nr

    E ρ(r)[ ] = Vext∫ (r)ρ(r)dr + F ρ(r)[ ]UNIVERSAL

    F ρ(r )[ ]= EKE ρ(r)[ ]+ EH ρ(r )[ ]+ EXC ρ(r)[ ]

    By inspection:

    Comparison to Hartree-Fock:

    •Formulates a expression for the unknown universal functional

    •The assumed form defines the exchange-correlation functional (which itself is still unknown)

    The Kohn-Sham approach is most widely used in practice (II).

    •The total charge density is written as a sum of charge densities of a set of fictitious, orthonormal orbitals

    •Giving rise to a set of non-linear equations that can be solved for the orbitals (again, eigenvalue/eigenfunction problem)

    •Two common approaches: self-consistency loop and direct diagonalization

    ρ(r) = ϕ i (r )2

    i=1

    N

  • DFT - the Kohn-Sham Equations

    H = − 1

    2∇2 −∑

    Znre−n

    +1re−e

    Hψ r1,..., rn( ) = Eψ r1,..., rn( )

    Non-interac+ngpart

    ContainsanElectron-Electron

    Distance!

    real Schrodinger Equation:- electron-electron interactions explicitly present - need to solve an interacting differential equation for a multi-

    dimensional function - at least its a linear equation

    instead, in Kohn-Sham DFT:

    ρ(r) = ϕ i (r )2

    i=1

    N

    electron kinetic energy

    electron-ion Coulomb

    interaction

    Hartree term. The electron-electron interaction expressed as the Coulomb interaction of a charge density interacting

    with itself

    exchange and correlation. our best guess at fixing everything wrong with the prior terms.

    Since the exact form of the exchange correlation functional not known, it needs to be approximated.

    The local density approximation (LDA) assumes that at each point in real space, the exchange -correlation energy at that point is equal to the exchange-correlation energy of a homogeneous electron gas of equivalent density.

    The generalized gradient approximations (GGAs) describes the exchange-correlation energy at a point in terms of the total density at that point and its gradient.

    Meta-GGAs use even higher order derivatives of the density at a point. e.g. SCAN

    Hybrid functionals - incorporation of exact exchange, to eliminate some of the self-interaction in LDA, GGA.

    van der Waals functionals - for weak interactions, layered materials, 2D material heterostructures.

    ~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*

    The uncertainty in the accuracy of DFT largely originates from the fact that we always need to select some form for this functional. Once the functional has been selected, we largely have to live with whatever physics comes out of it.

  • STRUCTURE AND TOTAL ENERGIES

    * can expect pretty good accuracy in the use of DFT for lattice constants, bulk moduli, etc. in comparison to experiment

    DFT is typically very useful for the prediction of ground state properties such as lattice constants, elastic moduli, and equation of state.

    experimental ao (A)

    devi

    atio

    n fr

    om e

    xp a

    o (A

    )

    Scheffler et al 2018 New J. Phys. 20 063020

  • DFT is typically very useful for the prediction of ground state properties such as lattice constants, elastic moduli, and equation of state.

    After HAAS, TRAN, AND BLAHA PHYSICAL

    REVIEW B 79, 085104 2009

    ~*~*~*~

    Can you find any systematic trends?

    DFT is typically very useful for the prediction of ground state properties such as lattice constants, elastic moduli, and equation of state.

    Scheffler et al 2018 New J. Phys. 20 063020

    SEMICONDUCTORS TRANSITION METAL CARBIDES & NITRIDES

    MAIN GROUP AND TRANSITION METALS IONIC CRYSTALS

  • PHASE STABILITIES

    * these can be trickier, because they rely on comparing total energies of different compounds to each other (rely on cancelation of errors)

    phase stability, e.g. CuGaTe2■ Thermodynamic equilibrium:

    Δ (eV)

    Δ (eV

    )

    GaCuTe

    ΔHf (CuGaTe2) = E(CuGaTe2) − (E(Cu(s)) + E(Ga(s)) + 2E(Te(s)))

    ■ Elemental references for chemical potentials:

    μCu = μoCu + ΔμCu μGa = μoGa + ΔμGa

    μTe = μoTe + ΔμTereference to solid elemental copper, shift from reference

    (variable)μoCu = EDFT(Cu(s))

    μCu + μGa + 2μTe = EDFT(CuGaTe2)

    ■ Combine:

    ΔμCu + ΔμGa + 2ΔμTe = ΔHf(CuGaTe2)

    where:

    ■ Size of the triangle set by the formation enthalpy; every point inside the triangle satisfies Eq. (2) by construction.

    (1a) (1b)

    (1c)

    (2)

  • ■ Need to avoid regions of phase space for which formation of competing compounds is favorable, e.g. for CuTe

    Δ (eV)

    Δ (eV

    )

    chemicalpotentialspace

    phase stability, e.g. CuGaTe2 GaCuTe

    ΔμCu + ΔμTe ≤ ΔHf(CuTe)e.g. for CuTe, require:

    Invariant points represent three-phase equilibria:

    Te, CuTe, CuGaTe2

    Cu, CuTe, CuGaTe2

    ■ Need to include all possible competing compounds

    phase stability, e.g. CuGaTe2 GaCuTe

    stability region

    Δ (eV)

    Δ (eV

    )

    chemicalpotentialspace

  • phase diagrams: composition and chemical potential

    ■ Can go back and forth between chemical potential and composition space. ■ Invariant points (three phase equilibria) in chemical potential space become

    areas in composition space; areas in chemical potential space become single points in composition space (NB: they do have a finite span in practice)

    stability region

    Δ (eV)

    Δ (eV

    )

    example:CuGaTe2

    compositionspace

    chemicalpotentialspaceGa

    Cu

    Te

    phase stability exampleCasestudyofchalcopyritematerialCuInSe2

    PBE HSE06 Experiment

    Kim, Pena-Martin, Rockett, Ertekin. Phys. Rev B 93 165202 (2016)

    ■ The hybrid functional improves the agreement with experiment.

    ■ Even so, DFT-computed PBE and HSE triangles are both too small: both methods underestimate the formation enthalpies of the chalcopyrite relative to experiment.

    ■ This means that the compound is underbound relative to elemental phases in DFT. We see this often. ■ Also note that both techniques do not get the correct phase boundaries

  • Elif Ertekin | [email protected] | Mechanical Science and Engineering

    example: spinels (AB2C4)

    23

    Energy Above Hull for Known Normal Spinels

    Freq

    uenc

    y

    0

    18

    35

    53

    70

    Energy Above Hull (eV/atom)

    0 (0.1,0.15] (0.25,0.3]

    • Distribution of formation energies about convex hull, pulled from Materials Project

    • These are all stable spinels, but DFT-PBE predicts roughly half to be unstable

    • Can be difficult to determine stability using only DFT-PBE

    insight: it is easier to predict instability than stability using conventional DFT

    FERE - Fitted Elemental Reference Energies

    Stevanovic, Lany, Zhang, and Zunger. Phys. Rev. B 85 115104 (2012).

    • Observation/Problem: Compound formation enthalpies are not very accurate in conventional DFT

    • Origin of Problem: Incomplete error cancellation between compounds and elements • Proposed solution: Fitted elemental reference energies (FERE)

    • introduce a shift for each elemental reference energy

    • solve for shifts that minimize root means squared error in computed formation energies across a set of compounds:

    • Improves thermodynamic boundary conditions for phase stabilities

    μFEREa = μDFTa + δμa

    ∑compounds

    (ΔHcomputedf − ΔHexpf )

    2

    δμa

    ΔHcomputedf = EDFT − ∑

    ana(μoa + δμa)

  • BAND STRUCTURES

    ■ Conventional DFT such as LDA and GGA tend to underestimate band gaps in generally non-systematic (unpredictable) ways

    ■ a result of the residual self-interaction error present in the exchange-correlation functional

    ■ Hybrid functionals improve the description of band gaps by incorporating some amount of exact exchange in the Kohn-Sham Hamiltonian to cancel out the self-interaction error

    ■ standard hybrid functionals (PBE0, HSE06), e.g. 25% exchange mixing

    ■ it is also possible to modify the amount of exchange mixing (be careful)

    ■ some formulations or recipes, such as choosing the amount of exact exchange based on the dielectric constant [see Gerosa et al., J. Phys.: Condens. Matter 30 (2018) 044003]

    ■ more computationally expensive ■ GW: many-body perturbation theory to obtain the self-

    interaction energy for quasi-particles ■ more computationally expensive, slow

    convergence, results sensitive to approach (i.e. single shot vs. self-consistent, vertex corrections, etc).

    Approaches to Band Structure

    Cu2O

    S. Lany, J. Phys. Cond. Mat. 27 283203 (2015)

    Chen W, Pasquarello A. Phys. Rev. B. 2012;

    86:035134

  • DEFECTS AND DOPING

    general formalism

    ■ Grandcanonicalthermodynamicformalismtoobtainformationenergiesofneutralandionizeddefects

    ■ Chargetransitionlevels:valuesoftheFermilevelatwhichthestablechargestateofthedefectchanges

    ■ EquilibriumFermilevelandcarrierconcentrationsaredeterminedfromchargeneutrality,

    ■ TherangeofachievableFermilevelsisindicatedbythedopabilitywindow

    FermiLevel

    DefectFormationEnergy

    VBM CBM

    q = 0donor

    acceptor

    q = -1q = 0

    q = +1

    charge transition levelsε(0 | − 1) ε(+1 |0)εeqFdopability window

  • early successes■ Negative-U model of DX centers in III-V semiconductors

    ● Chadi, D. J. & Chang, K. J. Theory of the Atomic and Electronic Structure of DX Centers in GaAs and AlxGa1-xAs Alloys. Phys. Rev. Lett. 61, 873–876 (1988).

    ● Jones, R. & Oberg, S. Structure and dynamics of the DX center in GaAs:Si. 44, 3407–3408 (1991).

    ■ Description of hydrogen in silicon (amphoteric) and other semiconductors ● Van de Walle, C. G., Denteneer, P. J. H., Bar--Yam, Y. & Pantelides, S. T. Theory of

    hydrogen diffusion and reactions in crystalline silicon. Phys. Rev. B 39, 10791–10808 (1989).

    ● Chang, K. J. & Chadi, D. J. Hydrogen bonding and diffusion in crystalline silicon. Phys. Rev. B 40, 11644–11653 (1989).

    ■ Identification of defects on surfaces ● Ebert, P. et al. Symmetric Versus Nonsymmetric Structure of the Phosphorus Vacancy in

    InP(110). Phys. Rev. Lett. 5816–5819 (2000). ● Zhang, S. B. & Zunger, A. Structure of the As Vacancies on GaAs(110) Surfaces. Phys.

    Rev. Lett. 77, 119–122 (1996). ● Kim, H. & Chelikowsky, J. R. Theoretical Scanning Tunneling Microscopy Images of the

    As Vacancy on the GaAs(110) Surface. Phys. Rev. Lett. 77, 1063–1066 (1996). ■ Identification of defects in new photovoltaic-active materials

    ● Zhang, S. B., Wei, S.--H. & Zunger, A. Defect physics of the CuInSe2 chalcopyrite semiconductor. 1–15 (1998).

    ■ Identification of defects in nitrides ● Van de Walle, C. G. First--principles calculations for defects and impurities:

    Applications to III--nitrides. J. Appl. Phys. 95, 3851 (2004).

    carrier concentrations etc.• equilibriumFermilevel:defectconcentrationsandelectronandholeconcentrationssimultaneouslysatisfychargeneutrality

    • dopabilitywindow:rangeofFermienergiesthatcouldbeaccessedbyincorporationofextrinsicdopants

    • example:showsapossibledonordefectthatpushestheFermienergyhighinthedopabilitywindow

    Intrinsicelectronconcentration:

    Intrinsicholeconcentration:

    n= 1V

    gc(E) fe(E)dEEc

    ∫ fe(E) =1

    1+ expE− EfkT

    ⎝⎜

    ⎠⎟

    fh(E) =1

    1+ expEf − EkT

    ⎝⎜

    ⎠⎟

    p = 1V

    gv(E) fh(E)dE∞

    EV

    concentrationofdefect



    CD ,q = N exp(−ΔHD ,qkBT

    )

    chargeneutrality


  • defect calculation framework

    Bulk Properties-------------------------------------• Compute energetics of main

    compound and competing compounds

    • Generate phase stability maps

    



    Perform Defect Calculations

    -------------------------------------- Identify native defects- Generate supercells and compute their energies

    Compute Formation Energies

    -------------------------------------- Calculate defect formation energies- Incorporate finite-size corrections



    Predict Carrier Concentrations

    -------------------------------------- calculate bulk DOS - use charge neutrality and

    statistical mechanics to obtain equilibrium Fermi level and carrier concentrations

    



    EFT

    norp

    How to Calculate & What Could Go Wrong?

    Typicalcomputationengines:• conventionalDFT–mostcommon;accuratebandgaps,bandedgepositions,andreferenceenergiesareaprerequisite

    • morerecently:• conventionalDFTwithsystematiccorrectionschemes

    • hybriddensityfunctionalDFT• GW• quantumMonteCarlo

    Traditionally,wecomputeenergiesusingDFTwithinthesupercellapproachwithperiodicboundaryconditions.

    Complicationsariselargelyfortworeasons:(i)accuracyofcomputedquantities,and(ii)finitesizeofsupercellssincewearegenerallyinterestedinthedilutedefectlimit.

    Rev.Mod.Phys.86,253(2014).

  • finite size effects

    Finite Size Effects

    + + + +

    + + + +

    + + + +

    + + + +

    + + + +

    • elasticfinitesizeeffectforanydefect

    • electrostaticfinitesizeeffectforchargeddefects(spuriouslystabilizeschargeddefects)

    • manywaystoestimate:• Leslie&Gillan(1985)• Makov-Payne(1995)• Freysoldt,Neugebauer,vandeWalle(2009)

    • Lany&Zunger(2008)

    ΔEf = (ED,q − Ehost )− niµi + q(EV + EF )i∑

    example: carrier concentrations in ternary diamond like semiconducting tellurides

    •strongly p-type material, across whole stability region

    •VCu-1 : killer defect that makes extrinsic n-type doping impossible

  • example: carrier concentrations in ternary diamond like semiconducting tellurides

    Summary

    ■ Density functional theory as the best available compromise today between accuracy attainable and computational cost; most practically well-suited for calculation of semiconducting solids

    ■ Useful for prediction of crystal structures, lattice constants, elastic constants

    ■ With some tweaks, can be made reasonably predictive for phase stability analysis

    ■ Band structure and band gaps: conventional DFT vs. hybrids vs. GW

    ■ Defects and dopability: defect formation energies, thermodynamic modeling, and carrier concentrations